[R] summary.manova and rank deficiency

[Peter Dalgaard]

"Simon Blomberg" <Simon.Blomberg at anu.edu.au> writes:

> Hi all,
> 
> I have received the following error from summary.manova:
> 
> Error in summary.manova(manova.test, test = "Pillai") : 
> 	residuals have rank 36 < 64
> 
> The data is simulated data for 64 variables. The design is a 2*2
> factorial with 10 replicates per treatment. Looking at the code for
> summary.manova, the error involves a problem with qr(). Does anyone
> have a suggestion as to how to deal with this error? The analysis
> works fine when there are only 32 variables.

So were you expecting "residuals have rank 36 < 32" in that case?? 

Some multivariate tests require that you can obtain a full rank
estimate of the covariance matrix, which requires the residual degrees
of freedom to be less than the number of response variables. Not sure
whether there is a meaningful way to proceed if that is not the case -
I suspect you'll get negative DF for the F tests and similar
awkwardness.  

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907